Adaptive Sensor Placement for Continuous Spaces

Authors: James Grant, Alexis Boukouvalas, Ryan-Rhys Griffiths, David Leslie, Sattar Vakili, Enrique Munoz De Cote

ICML 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In simulations we demonstrate our approach to have substantially lower and less variable regret than competitor algorithms.
Researcher Affiliation Collaboration 1PROWLER.io Ltd, Cambridge, United Kingdom 2STOR-i Centre for Doctoral Training, Lancaster University, Lancaster, United Kingdom 3Department of Physics, University of Cambridge, Cambridge, United Kingdom 4Department of Mathematics and Statistics, Lancaster University, Lancaster, United Kingdom.
Pseudocode Yes Algorithm 1 Thompson Sampling Inputs: Gamma prior parameters α, β > 0, upper truncation point λmax Iterative Phase: For t 1 For each k {1, . . . , Kt}, evaluate Hk,t(t 1) and Nk,t(t 1) and sample an index ψk,t TG(α+Hk,t(t 1), β+ t Nk,t(t 1), 0, λmax) Choose an action At At that maximises r(A) conditional on the true rate being given by the sampled ψk,t values, and observe the events in At
Open Source Code No The paper does not provide any explicit statement about releasing source code or a link to a code repository.
Open Datasets No The paper describes generating data through simulations based on rate functions but does not use a publicly available or open dataset for training, nor does it provide concrete access information for such a dataset.
Dataset Splits No The paper performs simulations but does not describe using standard training, validation, and test dataset splits for a pre-existing dataset.
Hardware Specification No The paper does not provide specific details about the hardware (e.g., GPU models, CPU types, memory, cloud instances) used for running the experiments.
Software Dependencies No The paper does not list specific software dependencies with version numbers (e.g., libraries, frameworks, or solvers).
Experiment Setup Yes Here, and throughout our experiments, we set the prior parameters for Thompson sampling to be α = 0.5 and β = 0.5/C, where scaling by cost C makes the prior relevant to the expected scale of costs in the problem. We also set the truncation λmax to be ten times the true maximal value of λ; λmax is an inconvenient parameter that is only needed for the theory, so we set it to a conservative large value that should have no influence on the real behaviour of the algorithm. The experiment is run 10 times for T = 1024 timesteps starting with K0 = 4 bins.